def meta_args_to_add(cls) -> [MetaArgument]:
return [
MetaArgument('cls_pbt_targets', Register.optimization_targets, allow_duplicates=True,
help_name='optimization target(s)'),
MetaArgument('cls_pbt_mutations', Register.pbt_mutations,
help_name='mutations to the copied checkpoint training state'),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
return super().meta_args_to_add() + [
MetaArgument('cls_network_stem', Register.network_stems, help_name='network stem', allowed_num=1, optional_for_loading=True),
MetaArgument('cls_network_heads', Register.network_heads, help_name='network heads', allow_duplicates=True, optional_for_loading=True),
MetaArgument('cls_network_cells', Register.network_cells, help_name='network cells', allow_duplicates=True),
MetaArgument('cls_network_cells_primitives', Register.primitive_sets, help_name='network cells primitives', allow_duplicates=True),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
kwargs = Register.get_my_kwargs(cls)
methods = Register.methods.filter_match_all(search=kwargs.get('search'))
return super().meta_args_to_add() + [
MetaArgument('cls_device', Register.devices_managers, help_name='device manager', allowed_num=1),
MetaArgument('cls_trainer', Register.trainers, help_name='trainer', allowed_num=1),
MetaArgument('cls_method', methods, help_name='method', allowed_num=1),
]
def meta_args_to_add(cls) -> [MetaArgument]:
return [
MetaArgument('cls_hpo_pymoo_sampler',
Register.hpo_pymoo_samplers,
help_name='sampling method',
allowed_num=1),
MetaArgument('cls_hpo_pymoo_crossover',
Register.hpo_pymoo_crossovers,
help_name='crossover method',
allowed_num=1),
MetaArgument('cls_hpo_pymoo_mutation',
Register.hpo_pymoo_mutations,
help_name='mutation method',
allowed_num=1),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
return super().meta_args_to_add() + [
MetaArgument('cls_network_body', Register.network_bodies, help_name='network', allowed_num=1),
]
def meta_args_to_add(cls, has_log_dict=True) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
callbacks = Register.training_callbacks
if not has_log_dict:
callbacks = callbacks.filter_match_all(requires_log_dict=False)
return super().meta_args_to_add() + [
MetaArgument('cls_callbacks',
callbacks,
help_name='training callbacks',
allow_duplicates=True),
MetaArgument('cls_exp_loggers',
Register.exp_loggers,
help_name='experiment logger',
allow_duplicates=True),
]
def meta_args_to_add(cls) -> [MetaArgument]:
return [
MetaArgument(
'cls_pbt_selector',
Register.pbt_selectors,
allowed_num=1,
help_name=
'selector to decide which checkpoints to keep and pursue'),
]
def meta_args_to_add(cls, estimator_filter: dict = None) -> [MetaArgument]:
estimators = Register.hpo_estimators
if isinstance(estimator_filter, dict):
estimators = estimators.filter_match_all(**estimator_filter)
return [
MetaArgument('cls_hpo_estimators',
estimators,
help_name='performance estimators',
allowed_num=(1, -1),
allow_duplicates=True),
MetaArgument('cls_hpo_pymoo_algorithm',
Register.hpo_pymoo_algorithms,
help_name='hyper-parameter algorithm',
allowed_num=1),
MetaArgument('cls_hpo_pymoo_termination',
Register.hpo_pymoo_terminators,
help_name='algorithm termination criteria',
allowed_num=1),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
kwargs = Register.get_my_kwargs(cls)
aug_sets = Register.augmentation_sets.filter_match_all(on_images=kwargs.get('images'))
return super().meta_args_to_add() + [
MetaArgument('cls_augmentations', aug_sets, help_name='data augmentation'),
]
def meta_args_to_add(cls, num_optimizers=1) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
return super().meta_args_to_add() + [
MetaArgument('cls_profile_fun',
Register.profile_functions,
help_name='profile function',
allowed_num=1),
]
def meta_args_to_add(cls,
estimator_filter: dict = None,
algorithm=True) -> [MetaArgument]:
estimators = Register.hpo_estimators
if isinstance(estimator_filter, dict):
estimators = estimators.filter_match_all(**estimator_filter)
meta = [
MetaArgument('cls_hpo_estimators',
estimators,
help_name='estimator',
allow_duplicates=True,
allowed_num=(1, -1))
]
if algorithm:
meta.append(
MetaArgument('cls_hpo_self_algorithm',
Register.hpo_self_algorithms,
help_name='hyper-parameter optimizer',
allowed_num=1))
return meta
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
networks = Register.networks.filter_match_all(search=True)
return super().meta_args_to_add() + [
MetaArgument('cls_device',
Register.devices_managers,
help_name='device manager',
allowed_num=1),
MetaArgument('cls_profiler',
Register.profilers,
help_name='profiler',
allowed_num=1),
MetaArgument('cls_data',
Register.data_sets,
help_name='data set',
allowed_num=1),
MetaArgument(
'cls_network', networks, help_name='network', allowed_num=1),
]
def meta_args_to_add(cls, algorithm=True) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
benchmark_sets = Register.benchmark_sets.filter_match_all(mini=True,
tabular=True)
return super().meta_args_to_add(algorithm=algorithm) + [
MetaArgument('cls_benchmark',
benchmark_sets,
allowed_num=1,
help_name='mini benchmark set to optimize on'),
]
def meta_args_to_add(cls, algorithm=True) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
benchmark_sets = Register.benchmark_sets.filter_match_all(mini=True,
tabular=True)
return super().meta_args_to_add(algorithm=algorithm) + [
MetaArgument(
'cls_benchmarks',
benchmark_sets,
help_name=
'optional benchmark sets, to evaluate specific architectures'),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
return super().meta_args_to_add() + [
MetaArgument(
'cls_benchmark',
Register.benchmark_sets,
allowed_num=(0, 1),
use_index=False,
help_name=
'immediately look up the search result in this benchmark set (optional)'
),
]
def meta_args_to_add(cls) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
benchmark_sets = Register.benchmark_sets.filter_match_all(mini=True,
tabular=True)
return super().meta_args_to_add() + [
MetaArgument(
'cls_benchmarks',
benchmark_sets,
allowed_num=(2, -1),
allow_duplicates=True,
help_name='mini benchmark sets to correlate with each other'),
] + SelfHPOUtils.meta_args_to_add(
algorithm=True, estimator_filter=dict(requires_bench=True))
def meta_args_to_add(cls, num_optimizers=1, search=True) -> [MetaArgument]:
"""
list meta arguments to add to argparse for when this class is chosen,
classes specified in meta arguments may have their own respective arguments
"""
kwargs = Register.get_my_kwargs(cls)
metrics = Register.metrics.filter_match_all(
distill=kwargs.get('distill'))
criteria = Register.criteria.filter_match_all(
distill=kwargs.get('distill'))
networks = Register.networks.filter_match_all(search=search)
return super().meta_args_to_add() + [
MetaArgument('cls_data',
Register.data_sets,
help_name='data set',
allowed_num=1),
MetaArgument(
'cls_network', networks, help_name='network', allowed_num=1),
MetaArgument('cls_criterion',
criteria,
help_name='criterion',
allowed_num=1),
MetaArgument('cls_metrics', metrics, help_name='training metric'),
MetaArgument('cls_initializers',
Register.initializers,
help_name='weight initializer'),
MetaArgument('cls_regularizers',
Register.regularizers,
help_name='regularizer'),
MetaArgument('cls_optimizers',
Register.optimizers,
help_name='optimizer',
allow_duplicates=True,
allowed_num=num_optimizers,
use_index=True),
MetaArgument('cls_schedulers',
Register.schedulers,
help_name='scheduler',
allow_duplicates=True,
allowed_num=(0, num_optimizers),
use_index=True),
]
def __init__(self, master, interactive: Interactive, node: GuiArgsTreeNode,
meta: MetaArgument):
self._config_enabled = False
super().__init__(master,
borderwidth=sizes.get('borderwidth'),
text="%s (allowed: %s)" %
(meta.argument.name, meta.limit_str()),
relief=None,
show=True)
self.interactive = interactive
self.node = node
self.meta = meta
self.children_widgets = []
self.str_add = '+'
self.str_rem = '-'
area_bottom = tk.Frame(master=self.sub_frame,
bg=colors.get('background_default'))
area_bottom.pack(side=tk.BOTTOM, expand=1, fill=tk.X)
self.var_rem_str = tk.StringVar(value=self.str_rem)
self.var_rem_str.trace_add('write', self._cb_rem_str)
self.rem_menu_button = tk.Menubutton(self.title_frame,
text=self.str_rem,
borderwidth=1,
relief="raised",
indicatoron=True)
self.rem_menu = tk.Menu(self.rem_menu_button, tearoff=False)
self.rem_menu_button.configure(menu=self.rem_menu)
self.rem_menu_button.pack(side=tk.RIGHT)
self.var_add_str = tk.StringVar(value=self.str_add)
self.var_add_str.trace_add('write', self._cb_add_str)
self.add_menu_button = tk.Menubutton(self.title_frame,
text=self.str_add,
borderwidth=1,
relief="raised",
indicatoron=True)
self.add_menu = tk.Menu(self.add_menu_button, tearoff=False)
self.add_menu_button.configure(menu=self.add_menu)
self.add_menu_button.pack(side=tk.RIGHT)